Hi. I am interested in knowing others experience with, and comments re, upscaling an already upscaled image using Gigapixel AI. For test, I upscaled an old scanned photo using Gigapixel. I then [re]upscaled the image created by previous upscale operation. I found that the “second pass” yielded much clearer results for facial details. Thoughts?
If I want 4x of a low resolution image, I typically do 2x then 2x. When you are dealing with pixels, it’s always “better” to reduce the amount of interpolation/sampling needed IMO.
If my image is already high resolution I may do something like 4x and then resize it by something like 0.5x.
In TVAI, I often do this with 720p video - taking it to 1440p then resizing it to 1080p after the fact with a simple unsharp mask applied.
Ultimately it depends what the content is and what algorithms are in use.
For me - if TVAI had the face recovery of Gigapixel, it would be a godsend on some kinds of very low res footage (yes I know you can export individual frames and do this… but it’s nice to not need extra manual steps).
Also I have found 6.2.x and 6.3.x to not give as good results on certain types of images/details as some late versions of 5.x. Ultimately Topaz are trying to create “one model fits all” (well, 5 or 6) but there are edge cases.
Do the workflow that works best for you!